Big data technology progress has led to improved precision of machine tool failure prediction. This paper utilises geometric data such as the structure and assembly relationships of CNC machine tools, combined with digital twin technology, to construct a multi-body kinematic/dynamic model. This enables simulation of the corresponding parts during machine tool operation, providing data support for predictive maintenance. Gray rough sets are introduced to optimise the BP neural network prediction algorithm. The predictive maintenance accuracy of CNC machine tools may be improved through the creation of an initial decision table, the use of grey relational analysis, and the processing of discrete information. The findings show that the BP neural-network prediction model based on grey rough sets has loss values of 0.099, 0.059 and 0.018 at three different iteration settings. In five different signal-to-noise ratio datasets, the loss function values are all less than 0.15, and the prediction accuracy exceeds 90%. This enables precise prediction and maintenance of CNC machine tool failures.
